notifier: prevent DEBUGASSERT recursion during system startup

[hujun260]

Summary

This PR fixes a critical recursion issue in the notifier system that occurs during early
system startup. The problem occurs when DEBUGASSERT is triggered in sched_lock(), which
internally calls panic_notifier_call_chain() - causing a cascading chain of recursive
DEBUGASSERT failures.

The fix adds explicit notifier head pointer checks before executing notifier chains,
preventing unnecessary lock operations on uninitialized notifiers during system startup
and error handling paths.

Problem Description

During early system startup or when DEBUGASSERT fails in the scheduler:

1. ARM64 fatal handler triggers DEBUGASSERT
2. _assert() is called
3. panic_notifier_call_chain() attempts lock acquisition
4. sched_lock() contains a DEBUGASSERT that checks rtcb && rtcb->lockcount < MAX_LOCK_COUNT
5. Since this_task() may return NULL during startup, the DEBUGASSERT fails again
6. Recursion cycle causes cascading failures instead of clean panic/reboot

Changes Made

Atomic Notifier Chain (ATOMIC_NOTIFIER_CALL macro)

Before:

do {
  FAR struct atomic_notifier_head *nh = (nhead);
  irqstate_t flags;
  flags = rspin_lock_irqsave_nopreempt(&nh->lock);
  notifier_call_chain(nh->head, (val), (v), -1, NULL);
  rspin_unlock_irqrestore_nopreempt(&nh->lock, flags);
} while(0)